drivers/block/xen-blkfront.c

   1 /*
   2  * blkfront.c
   3  *
   4  * XenLinux virtual block device driver.
   5  *
   6  * Copyright (c) 2003-2004, Keir Fraser & Steve Hand
   7  * Modifications by Mark A. Williamson are (c) Intel Research Cambridge
   8  * Copyright (c) 2004, Christian Limpach
   9  * Copyright (c) 2004, Andrew Warfield
  10  * Copyright (c) 2005, Christopher Clark
  11  * Copyright (c) 2005, XenSource Ltd
  12  *
  13  * This program is free software; you can redistribute it and/or
  14  * modify it under the terms of the GNU General Public License version 2
  15  * as published by the Free Software Foundation; or, when distributed
  16  * separately from the Linux kernel or incorporated into other
  17  * software packages, subject to the following license:
  18  *
  19  * Permission is hereby granted, free of charge, to any person obtaining a copy
  20  * of this source file (the "Software"), to deal in the Software without
  21  * restriction, including without limitation the rights to use, copy, modify,
  22  * merge, publish, distribute, sublicense, and/or sell copies of the Software,
  23  * and to permit persons to whom the Software is furnished to do so, subject to
  24  * the following conditions:
  25  *
  26  * The above copyright notice and this permission notice shall be included in
  27  * all copies or substantial portions of the Software.
  28  *
  29  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  30  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  31  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  32  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  33  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  34  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  35  * IN THE SOFTWARE.
  36  */
  37
  38 #include <linux/interrupt.h>
  39 #include <linux/blkdev.h>
  40 #include <linux/hdreg.h>
  41 #include <linux/cdrom.h>
  42 #include <linux/module.h>
  43
  44 #include <xen/xenbus.h>
  45 #include <xen/grant_table.h>
  46 #include <xen/events.h>
  47 #include <xen/page.h>
  48
  49 #include <xen/interface/grant_table.h>
  50 #include <xen/interface/io/blkif.h>
  51 #include <xen/interface/io/protocols.h>
  52
  53 #include <asm/xen/hypervisor.h>
  54
  55 enum blkif_state {
  56         BLKIF_STATE_DISCONNECTED,
  57         BLKIF_STATE_CONNECTED,
  58         BLKIF_STATE_SUSPENDED,
  59 };
  60
  61 struct blk_shadow {
  62         struct blkif_request req;
  63         unsigned long request;
  64         unsigned long frame[BLKIF_MAX_SEGMENTS_PER_REQUEST];
  65 };
  66
  67 static struct block_device_operations xlvbd_block_fops;
  68
  69 #define BLK_RING_SIZE __RING_SIZE((struct blkif_sring *)0, PAGE_SIZE)
  70
  71 /*
  72  * We have one of these per vbd, whether ide, scsi or 'other'.  They
  73  * hang in private_data off the gendisk structure. We may end up
  74  * putting all kinds of interesting stuff here :-)
  75  */
  76 struct blkfront_info
  77 {
  78         struct xenbus_device *xbdev;
  79         struct gendisk *gd;
  80         int vdevice;
  81         blkif_vdev_t handle;
  82         enum blkif_state connected;
  83         int ring_ref;
  84         struct blkif_front_ring ring;
  85         unsigned int evtchn, irq;
  86         struct request_queue *rq;
  87         struct work_struct work;
  88         struct gnttab_free_callback callback;
  89         struct blk_shadow shadow[BLK_RING_SIZE];
  90         unsigned long shadow_free;
  91         int feature_barrier;
  92         int is_ready;
  93
  94         /**
  95          * The number of people holding this device open.  We won't allow a
  96          * hot-unplug unless this is 0.
  97          */
  98         int users;
  99 };
 100
 101 static DEFINE_SPINLOCK(blkif_io_lock);
 102
 103 #define MAXIMUM_OUTSTANDING_BLOCK_REQS \
 104         (BLKIF_MAX_SEGMENTS_PER_REQUEST * BLK_RING_SIZE)
 105 #define GRANT_INVALID_REF       0
 106
 107 #define PARTS_PER_DISK          16
 108
 109 #define BLKIF_MAJOR(dev) ((dev)>>8)
 110 #define BLKIF_MINOR(dev) ((dev) & 0xff)
 111
 112 #define DEV_NAME        "xvd"   /* name in /dev */
 113
 114 /* Information about our VBDs. */
 115 #define MAX_VBDS 64
 116 static LIST_HEAD(vbds_list);
 117
 118 static int get_id_from_freelist(struct blkfront_info *info)
 119 {
 120         unsigned long free = info->shadow_free;
 121         BUG_ON(free > BLK_RING_SIZE);
 122         info->shadow_free = info->shadow[free].req.id;
 123         info->shadow[free].req.id = 0x0fffffee; /* debug */
 124         return free;
 125 }
 126
 127 static void add_id_to_freelist(struct blkfront_info *info,
 128                                unsigned long id)
 129 {
 130         info->shadow[id].req.id  = info->shadow_free;
 131         info->shadow[id].request = 0;
 132         info->shadow_free = id;
 133 }
 134
 135 static void blkif_restart_queue_callback(void *arg)
 136 {
 137         struct blkfront_info *info = (struct blkfront_info *)arg;
 138         schedule_work(&info->work);
 139 }
 140
 141 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
 142 {
 143         /* We don't have real geometry info, but let's at least return
 144            values consistent with the size of the device */
 145         sector_t nsect = get_capacity(bd->bd_disk);
 146         sector_t cylinders = nsect;
 147
 148         hg->heads = 0xff;
 149         hg->sectors = 0x3f;
 150         sector_div(cylinders, hg->heads * hg->sectors);
 151         hg->cylinders = cylinders;
 152         if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
 153                 hg->cylinders = 0xffff;
 154         return 0;
 155 }
 156
 157 int blkif_ioctl(struct inode *inode, struct file *filep,
 158                 unsigned command, unsigned long argument)
 159 {
 160         struct blkfront_info *info =
 161                 inode->i_bdev->bd_disk->private_data;
 162         int i;
 163
 164         dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
 165                 command, (long)argument);
 166
 167         switch (command) {
 168         case CDROMMULTISESSION:
 169                 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
 170                 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
 171                         if (put_user(0, (char __user *)(argument + i)))
 172                                 return -EFAULT;
 173                 return 0;
 174
 175         case CDROM_GET_CAPABILITY: {
 176                 struct gendisk *gd = info->gd;
 177                 if (gd->flags & GENHD_FL_CD)
 178                         return 0;
 179                 return -EINVAL;
 180         }
 181
 182         default:
 183                 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
 184                   command);*/
 185                 return -EINVAL; /* same return as native Linux */
 186         }
 187
 188         return 0;
 189 }
 190
 191 /*
 192  * blkif_queue_request
 193  *
 194  * request block io
 195  *
 196  * id: for guest use only.
 197  * operation: BLKIF_OP_{READ,WRITE,PROBE}
 198  * buffer: buffer to read/write into. this should be a
 199  *   virtual address in the guest os.
 200  */
 201 static int blkif_queue_request(struct request *req)
 202 {
 203         struct blkfront_info *info = req->rq_disk->private_data;
 204         unsigned long buffer_mfn;
 205         struct blkif_request *ring_req;
 206         struct req_iterator iter;
 207         struct bio_vec *bvec;
 208         unsigned long id;
 209         unsigned int fsect, lsect;
 210         int ref;
 211         grant_ref_t gref_head;
 212
 213         if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
 214                 return 1;
 215
 216         if (gnttab_alloc_grant_references(
 217                 BLKIF_MAX_SEGMENTS_PER_REQUEST, &gref_head) < 0) {
 218                 gnttab_request_free_callback(
 219                         &info->callback,
 220                         blkif_restart_queue_callback,
 221                         info,
 222                         BLKIF_MAX_SEGMENTS_PER_REQUEST);
 223                 return 1;
 224         }
 225
 226         /* Fill out a communications ring structure. */
 227         ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
 228         id = get_id_from_freelist(info);
 229         info->shadow[id].request = (unsigned long)req;
 230
 231         ring_req->id = id;
 232         ring_req->sector_number = (blkif_sector_t)req->sector;
 233         ring_req->handle = info->handle;
 234
 235         ring_req->operation = rq_data_dir(req) ?
 236                 BLKIF_OP_WRITE : BLKIF_OP_READ;
 237         if (blk_barrier_rq(req))
 238                 ring_req->operation = BLKIF_OP_WRITE_BARRIER;
 239
 240         ring_req->nr_segments = 0;
 241         rq_for_each_segment(bvec, req, iter) {
 242                 BUG_ON(ring_req->nr_segments == BLKIF_MAX_SEGMENTS_PER_REQUEST);
 243                 buffer_mfn = pfn_to_mfn(page_to_pfn(bvec->bv_page));
 244                 fsect = bvec->bv_offset >> 9;
 245                 lsect = fsect + (bvec->bv_len >> 9) - 1;
 246                 /* install a grant reference. */
 247                 ref = gnttab_claim_grant_reference(&gref_head);
 248                 BUG_ON(ref == -ENOSPC);
 249
 250                 gnttab_grant_foreign_access_ref(
 251                                 ref,
 252                                 info->xbdev->otherend_id,
 253                                 buffer_mfn,
 254                                 rq_data_dir(req) );
 255
 256                 info->shadow[id].frame[ring_req->nr_segments] =
 257                                 mfn_to_pfn(buffer_mfn);
 258
 259                 ring_req->seg[ring_req->nr_segments] =
 260                                 (struct blkif_request_segment) {
 261                                         .gref       = ref,
 262                                         .first_sect = fsect,
 263                                         .last_sect  = lsect };
 264
 265                 ring_req->nr_segments++;
 266         }
 267
 268         info->ring.req_prod_pvt++;
 269
 270         /* Keep a private copy so we can reissue requests when recovering. */
 271         info->shadow[id].req = *ring_req;
 272
 273         gnttab_free_grant_references(gref_head);
 274
 275         return 0;
 276 }
 277
 278
 279 static inline void flush_requests(struct blkfront_info *info)
 280 {
 281         int notify;
 282
 283         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify);
 284
 285         if (notify)
 286                 notify_remote_via_irq(info->irq);
 287 }
 288
 289 /*
 290  * do_blkif_request
 291  *  read a block; request is in a request queue
 292  */
 293 static void do_blkif_request(struct request_queue *rq)
 294 {
 295         struct blkfront_info *info = NULL;
 296         struct request *req;
 297         int queued;
 298
 299         pr_debug("Entered do_blkif_request\n");
 300
 301         queued = 0;
 302
 303         while ((req = elv_next_request(rq)) != NULL) {
 304                 info = req->rq_disk->private_data;
 305                 if (!blk_fs_request(req)) {
 306                         end_request(req, 0);
 307                         continue;
 308                 }
 309
 310                 if (RING_FULL(&info->ring))
 311                         goto wait;
 312
 313                 pr_debug("do_blk_req %p: cmd %p, sec %lx, "
 314                          "(%u/%li) buffer:%p [%s]\n",
 315                          req, req->cmd, (unsigned long)req->sector,
 316                          req->current_nr_sectors,
 317                          req->nr_sectors, req->buffer,
 318                          rq_data_dir(req) ? "write" : "read");
 319
 320
 321                 blkdev_dequeue_request(req);
 322                 if (blkif_queue_request(req)) {
 323                         blk_requeue_request(rq, req);
 324 wait:
 325                         /* Avoid pointless unplugs. */
 326                         blk_stop_queue(rq);
 327                         break;
 328                 }
 329
 330                 queued++;
 331         }
 332
 333         if (queued != 0)
 334                 flush_requests(info);
 335 }
 336
 337 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size)
 338 {
 339         struct request_queue *rq;
 340
 341         rq = blk_init_queue(do_blkif_request, &blkif_io_lock);
 342         if (rq == NULL)
 343                 return -1;
 344
 345         elevator_init(rq, "noop");
 346
 347         /* Hard sector size and max sectors impersonate the equiv. hardware. */
 348         blk_queue_hardsect_size(rq, sector_size);
 349         blk_queue_max_sectors(rq, 512);
 350
 351         /* Each segment in a request is up to an aligned page in size. */
 352         blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
 353         blk_queue_max_segment_size(rq, PAGE_SIZE);
 354
 355         /* Ensure a merged request will fit in a single I/O ring slot. */
 356         blk_queue_max_phys_segments(rq, BLKIF_MAX_SEGMENTS_PER_REQUEST);
 357         blk_queue_max_hw_segments(rq, BLKIF_MAX_SEGMENTS_PER_REQUEST);
 358
 359         /* Make sure buffer addresses are sector-aligned. */
 360         blk_queue_dma_alignment(rq, 511);
 361
 362         /* Make sure we don't use bounce buffers. */
 363         blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
 364
 365         gd->queue = rq;
 366
 367         return 0;
 368 }
 369
 370
 371 static int xlvbd_barrier(struct blkfront_info *info)
 372 {
 373         int err;
 374
 375         err = blk_queue_ordered(info->rq,
 376                                 info->feature_barrier ? QUEUE_ORDERED_DRAIN : QUEUE_ORDERED_NONE,
 377                                 NULL);
 378
 379         if (err)
 380                 return err;
 381
 382         printk(KERN_INFO "blkfront: %s: barriers %s\n",
 383                info->gd->disk_name,
 384                info->feature_barrier ? "enabled" : "disabled");
 385         return 0;
 386 }
 387
 388
 389 static int xlvbd_alloc_gendisk(int minor, blkif_sector_t capacity,
 390                                int vdevice, u16 vdisk_info, u16 sector_size,
 391                                struct blkfront_info *info)
 392 {
 393         struct gendisk *gd;
 394         int nr_minors = 1;
 395         int err = -ENODEV;
 396
 397         BUG_ON(info->gd != NULL);
 398         BUG_ON(info->rq != NULL);
 399
 400         if ((minor % PARTS_PER_DISK) == 0)
 401                 nr_minors = PARTS_PER_DISK;
 402
 403         gd = alloc_disk(nr_minors);
 404         if (gd == NULL)
 405                 goto out;
 406
 407         if (nr_minors > 1)
 408                 sprintf(gd->disk_name, "%s%c", DEV_NAME,
 409                         'a' + minor / PARTS_PER_DISK);
 410         else
 411                 sprintf(gd->disk_name, "%s%c%d", DEV_NAME,
 412                         'a' + minor / PARTS_PER_DISK,
 413                         minor % PARTS_PER_DISK);
 414
 415         gd->major = XENVBD_MAJOR;
 416         gd->first_minor = minor;
 417         gd->fops = &xlvbd_block_fops;
 418         gd->private_data = info;
 419         gd->driverfs_dev = &(info->xbdev->dev);
 420         set_capacity(gd, capacity);
 421
 422         if (xlvbd_init_blk_queue(gd, sector_size)) {
 423                 del_gendisk(gd);
 424                 goto out;
 425         }
 426
 427         info->rq = gd->queue;
 428         info->gd = gd;
 429
 430         if (info->feature_barrier)
 431                 xlvbd_barrier(info);
 432
 433         if (vdisk_info & VDISK_READONLY)
 434                 set_disk_ro(gd, 1);
 435
 436         if (vdisk_info & VDISK_REMOVABLE)
 437                 gd->flags |= GENHD_FL_REMOVABLE;
 438
 439         if (vdisk_info & VDISK_CDROM)
 440                 gd->flags |= GENHD_FL_CD;
 441
 442         return 0;
 443
 444  out:
 445         return err;
 446 }
 447
 448 static void kick_pending_request_queues(struct blkfront_info *info)
 449 {
 450         if (!RING_FULL(&info->ring)) {
 451                 /* Re-enable calldowns. */
 452                 blk_start_queue(info->rq);
 453                 /* Kick things off immediately. */
 454                 do_blkif_request(info->rq);
 455         }
 456 }
 457
 458 static void blkif_restart_queue(struct work_struct *work)
 459 {
 460         struct blkfront_info *info = container_of(work, struct blkfront_info, work);
 461
 462         spin_lock_irq(&blkif_io_lock);
 463         if (info->connected == BLKIF_STATE_CONNECTED)
 464                 kick_pending_request_queues(info);
 465         spin_unlock_irq(&blkif_io_lock);
 466 }
 467
 468 static void blkif_free(struct blkfront_info *info, int suspend)
 469 {
 470         /* Prevent new requests being issued until we fix things up. */
 471         spin_lock_irq(&blkif_io_lock);
 472         info->connected = suspend ?
 473                 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
 474         /* No more blkif_request(). */
 475         if (info->rq)
 476                 blk_stop_queue(info->rq);
 477         /* No more gnttab callback work. */
 478         gnttab_cancel_free_callback(&info->callback);
 479         spin_unlock_irq(&blkif_io_lock);
 480
 481         /* Flush gnttab callback work. Must be done with no locks held. */
 482         flush_scheduled_work();
 483
 484         /* Free resources associated with old device channel. */
 485         if (info->ring_ref != GRANT_INVALID_REF) {
 486                 gnttab_end_foreign_access(info->ring_ref, 0,
 487                                           (unsigned long)info->ring.sring);
 488                 info->ring_ref = GRANT_INVALID_REF;
 489                 info->ring.sring = NULL;
 490         }
 491         if (info->irq)
 492                 unbind_from_irqhandler(info->irq, info);
 493         info->evtchn = info->irq = 0;
 494
 495 }
 496
 497 static void blkif_completion(struct blk_shadow *s)
 498 {
 499         int i;
 500         for (i = 0; i < s->req.nr_segments; i++)
 501                 gnttab_end_foreign_access(s->req.seg[i].gref, 0, 0UL);
 502 }
 503
 504 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
 505 {
 506         struct request *req;
 507         struct blkif_response *bret;
 508         RING_IDX i, rp;
 509         unsigned long flags;
 510         struct blkfront_info *info = (struct blkfront_info *)dev_id;
 511         int error;
 512
 513         spin_lock_irqsave(&blkif_io_lock, flags);
 514
 515         if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
 516                 spin_unlock_irqrestore(&blkif_io_lock, flags);
 517                 return IRQ_HANDLED;
 518         }
 519
 520  again:
 521         rp = info->ring.sring->rsp_prod;
 522         rmb(); /* Ensure we see queued responses up to 'rp'. */
 523
 524         for (i = info->ring.rsp_cons; i != rp; i++) {
 525                 unsigned long id;
 526                 int ret;
 527
 528                 bret = RING_GET_RESPONSE(&info->ring, i);
 529                 id   = bret->id;
 530                 req  = (struct request *)info->shadow[id].request;
 531
 532                 blkif_completion(&info->shadow[id]);
 533
 534                 add_id_to_freelist(info, id);
 535
 536                 error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
 537                 switch (bret->operation) {
 538                 case BLKIF_OP_WRITE_BARRIER:
 539                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
 540                                 printk(KERN_WARNING "blkfront: %s: write barrier op failed\n",
 541                                        info->gd->disk_name);
 542                                 error = -EOPNOTSUPP;
 543                                 info->feature_barrier = 0;
 544                                 xlvbd_barrier(info);
 545                         }
 546                         /* fall through */
 547                 case BLKIF_OP_READ:
 548                 case BLKIF_OP_WRITE:
 549                         if (unlikely(bret->status != BLKIF_RSP_OKAY))
 550                                 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
 551                                         "request: %x\n", bret->status);
 552
 553                         ret = __blk_end_request(req, error, blk_rq_bytes(req));
 554                         BUG_ON(ret);
 555                         break;
 556                 default:
 557                         BUG();
 558                 }
 559         }
 560
 561         info->ring.rsp_cons = i;
 562
 563         if (i != info->ring.req_prod_pvt) {
 564                 int more_to_do;
 565                 RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
 566                 if (more_to_do)
 567                         goto again;
 568         } else
 569                 info->ring.sring->rsp_event = i + 1;
 570
 571         kick_pending_request_queues(info);
 572
 573         spin_unlock_irqrestore(&blkif_io_lock, flags);
 574
 575         return IRQ_HANDLED;
 576 }
 577
 578
 579 static int setup_blkring(struct xenbus_device *dev,
 580                          struct blkfront_info *info)
 581 {
 582         struct blkif_sring *sring;
 583         int err;
 584
 585         info->ring_ref = GRANT_INVALID_REF;
 586
 587         sring = (struct blkif_sring *)__get_free_page(GFP_NOIO | __GFP_HIGH);
 588         if (!sring) {
 589                 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
 590                 return -ENOMEM;
 591         }
 592         SHARED_RING_INIT(sring);
 593         FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
 594
 595         err = xenbus_grant_ring(dev, virt_to_mfn(info->ring.sring));
 596         if (err < 0) {
 597                 free_page((unsigned long)sring);
 598                 info->ring.sring = NULL;
 599                 goto fail;
 600         }
 601         info->ring_ref = err;
 602
 603         err = xenbus_alloc_evtchn(dev, &info->evtchn);
 604         if (err)
 605                 goto fail;
 606
 607         err = bind_evtchn_to_irqhandler(info->evtchn,
 608                                         blkif_interrupt,
 609                                         IRQF_SAMPLE_RANDOM, "blkif", info);
 610         if (err <= 0) {
 611                 xenbus_dev_fatal(dev, err,
 612                                  "bind_evtchn_to_irqhandler failed");
 613                 goto fail;
 614         }
 615         info->irq = err;
 616
 617         return 0;
 618 fail:
 619         blkif_free(info, 0);
 620         return err;
 621 }
 622
 623
 624 /* Common code used when first setting up, and when resuming. */
 625 static int talk_to_backend(struct xenbus_device *dev,
 626                            struct blkfront_info *info)
 627 {
 628         const char *message = NULL;
 629         struct xenbus_transaction xbt;
 630         int err;
 631
 632         /* Create shared ring, alloc event channel. */
 633         err = setup_blkring(dev, info);
 634         if (err)
 635                 goto out;
 636
 637 again:
 638         err = xenbus_transaction_start(&xbt);
 639         if (err) {
 640                 xenbus_dev_fatal(dev, err, "starting transaction");
 641                 goto destroy_blkring;
 642         }
 643
 644         err = xenbus_printf(xbt, dev->nodename,
 645                             "ring-ref", "%u", info->ring_ref);
 646         if (err) {
 647                 message = "writing ring-ref";
 648                 goto abort_transaction;
 649         }
 650         err = xenbus_printf(xbt, dev->nodename,
 651                             "event-channel", "%u", info->evtchn);
 652         if (err) {
 653                 message = "writing event-channel";
 654                 goto abort_transaction;
 655         }
 656         err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
 657                             XEN_IO_PROTO_ABI_NATIVE);
 658         if (err) {
 659                 message = "writing protocol";
 660                 goto abort_transaction;
 661         }
 662
 663         err = xenbus_transaction_end(xbt, 0);
 664         if (err) {
 665                 if (err == -EAGAIN)
 666                         goto again;
 667                 xenbus_dev_fatal(dev, err, "completing transaction");
 668                 goto destroy_blkring;
 669         }
 670
 671         xenbus_switch_state(dev, XenbusStateInitialised);
 672
 673         return 0;
 674
 675  abort_transaction:
 676         xenbus_transaction_end(xbt, 1);
 677         if (message)
 678                 xenbus_dev_fatal(dev, err, "%s", message);
 679  destroy_blkring:
 680         blkif_free(info, 0);
 681  out:
 682         return err;
 683 }
 684
 685
 686 /**
 687  * Entry point to this code when a new device is created.  Allocate the basic
 688  * structures and the ring buffer for communication with the backend, and
 689  * inform the backend of the appropriate details for those.  Switch to
 690  * Initialised state.
 691  */
 692 static int blkfront_probe(struct xenbus_device *dev,
 693                           const struct xenbus_device_id *id)
 694 {
 695         int err, vdevice, i;
 696         struct blkfront_info *info;
 697
 698         /* FIXME: Use dynamic device id if this is not set. */
 699         err = xenbus_scanf(XBT_NIL, dev->nodename,
 700                            "virtual-device", "%i", &vdevice);
 701         if (err != 1) {
 702                 xenbus_dev_fatal(dev, err, "reading virtual-device");
 703                 return err;
 704         }
 705
 706         info = kzalloc(sizeof(*info), GFP_KERNEL);
 707         if (!info) {
 708                 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
 709                 return -ENOMEM;
 710         }
 711
 712         info->xbdev = dev;
 713         info->vdevice = vdevice;
 714         info->connected = BLKIF_STATE_DISCONNECTED;
 715         INIT_WORK(&info->work, blkif_restart_queue);
 716
 717         for (i = 0; i < BLK_RING_SIZE; i++)
 718                 info->shadow[i].req.id = i+1;
 719         info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
 720
 721         /* Front end dir is a number, which is used as the id. */
 722         info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
 723         dev->dev.driver_data = info;
 724
 725         err = talk_to_backend(dev, info);
 726         if (err) {
 727                 kfree(info);
 728                 dev->dev.driver_data = NULL;
 729                 return err;
 730         }
 731
 732         return 0;
 733 }
 734
 735
 736 static int blkif_recover(struct blkfront_info *info)
 737 {
 738         int i;
 739         struct blkif_request *req;
 740         struct blk_shadow *copy;
 741         int j;
 742
 743         /* Stage 1: Make a safe copy of the shadow state. */
 744         copy = kmalloc(sizeof(info->shadow),
 745                        GFP_NOIO | __GFP_REPEAT | __GFP_HIGH);
 746         if (!copy)
 747                 return -ENOMEM;
 748         memcpy(copy, info->shadow, sizeof(info->shadow));
 749
 750         /* Stage 2: Set up free list. */
 751         memset(&info->shadow, 0, sizeof(info->shadow));
 752         for (i = 0; i < BLK_RING_SIZE; i++)
 753                 info->shadow[i].req.id = i+1;
 754         info->shadow_free = info->ring.req_prod_pvt;
 755         info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
 756
 757         /* Stage 3: Find pending requests and requeue them. */
 758         for (i = 0; i < BLK_RING_SIZE; i++) {
 759                 /* Not in use? */
 760                 if (copy[i].request == 0)
 761                         continue;
 762
 763                 /* Grab a request slot and copy shadow state into it. */
 764                 req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
 765                 *req = copy[i].req;
 766
 767                 /* We get a new request id, and must reset the shadow state. */
 768                 req->id = get_id_from_freelist(info);
 769                 memcpy(&info->shadow[req->id], &copy[i], sizeof(copy[i]));
 770
 771                 /* Rewrite any grant references invalidated by susp/resume. */
 772                 for (j = 0; j < req->nr_segments; j++)
 773                         gnttab_grant_foreign_access_ref(
 774                                 req->seg[j].gref,
 775                                 info->xbdev->otherend_id,
 776                                 pfn_to_mfn(info->shadow[req->id].frame[j]),
 777                                 rq_data_dir(
 778                                         (struct request *)
 779                                         info->shadow[req->id].request));
 780                 info->shadow[req->id].req = *req;
 781
 782                 info->ring.req_prod_pvt++;
 783         }
 784
 785         kfree(copy);
 786
 787         xenbus_switch_state(info->xbdev, XenbusStateConnected);
 788
 789         spin_lock_irq(&blkif_io_lock);
 790
 791         /* Now safe for us to use the shared ring */
 792         info->connected = BLKIF_STATE_CONNECTED;
 793
 794         /* Send off requeued requests */
 795         flush_requests(info);
 796
 797         /* Kick any other new requests queued since we resumed */
 798         kick_pending_request_queues(info);
 799
 800         spin_unlock_irq(&blkif_io_lock);
 801
 802         return 0;
 803 }
 804
 805 /**
 806  * We are reconnecting to the backend, due to a suspend/resume, or a backend
 807  * driver restart.  We tear down our blkif structure and recreate it, but
 808  * leave the device-layer structures intact so that this is transparent to the
 809  * rest of the kernel.
 810  */
 811 static int blkfront_resume(struct xenbus_device *dev)
 812 {
 813         struct blkfront_info *info = dev->dev.driver_data;
 814         int err;
 815
 816         dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
 817
 818         blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
 819
 820         err = talk_to_backend(dev, info);
 821         if (info->connected == BLKIF_STATE_SUSPENDED && !err)
 822                 err = blkif_recover(info);
 823
 824         return err;
 825 }
 826
 827
 828 /*
 829  * Invoked when the backend is finally 'ready' (and has told produced
 830  * the details about the physical device - #sectors, size, etc).
 831  */
 832 static void blkfront_connect(struct blkfront_info *info)
 833 {
 834         unsigned long long sectors;
 835         unsigned long sector_size;
 836         unsigned int binfo;
 837         int err;
 838
 839         if ((info->connected == BLKIF_STATE_CONNECTED) ||
 840             (info->connected == BLKIF_STATE_SUSPENDED) )
 841                 return;
 842
 843         dev_dbg(&info->xbdev->dev, "%s:%s.\n",
 844                 __func__, info->xbdev->otherend);
 845
 846         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
 847                             "sectors", "%llu", &sectors,
 848                             "info", "%u", &binfo,
 849                             "sector-size", "%lu", &sector_size,
 850                             NULL);
 851         if (err) {
 852                 xenbus_dev_fatal(info->xbdev, err,
 853                                  "reading backend fields at %s",
 854                                  info->xbdev->otherend);
 855                 return;
 856         }
 857
 858         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
 859                             "feature-barrier", "%lu", &info->feature_barrier,
 860                             NULL);
 861         if (err)
 862                 info->feature_barrier = 0;
 863
 864         err = xlvbd_alloc_gendisk(BLKIF_MINOR(info->vdevice),
 865                                   sectors, info->vdevice,
 866                                   binfo, sector_size, info);
 867         if (err) {
 868                 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
 869                                  info->xbdev->otherend);
 870                 return;
 871         }
 872
 873         xenbus_switch_state(info->xbdev, XenbusStateConnected);
 874
 875         /* Kick pending requests. */
 876         spin_lock_irq(&blkif_io_lock);
 877         info->connected = BLKIF_STATE_CONNECTED;
 878         kick_pending_request_queues(info);
 879         spin_unlock_irq(&blkif_io_lock);
 880
 881         add_disk(info->gd);
 882
 883         info->is_ready = 1;
 884 }
 885
 886 /**
 887  * Handle the change of state of the backend to Closing.  We must delete our
 888  * device-layer structures now, to ensure that writes are flushed through to
 889  * the backend.  Once is this done, we can switch to Closed in
 890  * acknowledgement.
 891  */
 892 static void blkfront_closing(struct xenbus_device *dev)
 893 {
 894         struct blkfront_info *info = dev->dev.driver_data;
 895         unsigned long flags;
 896
 897         dev_dbg(&dev->dev, "blkfront_closing: %s removed\n", dev->nodename);
 898
 899         if (info->rq == NULL)
 900                 goto out;
 901
 902         spin_lock_irqsave(&blkif_io_lock, flags);
 903
 904         del_gendisk(info->gd);
 905
 906         /* No more blkif_request(). */
 907         blk_stop_queue(info->rq);
 908
 909         /* No more gnttab callback work. */
 910         gnttab_cancel_free_callback(&info->callback);
 911         spin_unlock_irqrestore(&blkif_io_lock, flags);
 912
 913         /* Flush gnttab callback work. Must be done with no locks held. */
 914         flush_scheduled_work();
 915
 916         blk_cleanup_queue(info->rq);
 917         info->rq = NULL;
 918
 919  out:
 920         xenbus_frontend_closed(dev);
 921 }
 922
 923 /**
 924  * Callback received when the backend's state changes.
 925  */
 926 static void backend_changed(struct xenbus_device *dev,
 927                             enum xenbus_state backend_state)
 928 {
 929         struct blkfront_info *info = dev->dev.driver_data;
 930         struct block_device *bd;
 931
 932         dev_dbg(&dev->dev, "blkfront:backend_changed.\n");
 933
 934         switch (backend_state) {
 935         case XenbusStateInitialising:
 936         case XenbusStateInitWait:
 937         case XenbusStateInitialised:
 938         case XenbusStateUnknown:
 939         case XenbusStateClosed:
 940                 break;
 941
 942         case XenbusStateConnected:
 943                 blkfront_connect(info);
 944                 break;
 945
 946         case XenbusStateClosing:
 947                 bd = bdget_disk(info->gd, 0);
 948                 if (bd == NULL)
 949                         xenbus_dev_fatal(dev, -ENODEV, "bdget failed");
 950
 951                 mutex_lock(&bd->bd_mutex);
 952                 if (info->users > 0)
 953                         xenbus_dev_error(dev, -EBUSY,
 954                                          "Device in use; refusing to close");
 955                 else
 956                         blkfront_closing(dev);
 957                 mutex_unlock(&bd->bd_mutex);
 958                 bdput(bd);
 959                 break;
 960         }
 961 }
 962
 963 static int blkfront_remove(struct xenbus_device *dev)
 964 {
 965         struct blkfront_info *info = dev->dev.driver_data;
 966
 967         dev_dbg(&dev->dev, "blkfront_remove: %s removed\n", dev->nodename);
 968
 969         blkif_free(info, 0);
 970
 971         kfree(info);
 972
 973         return 0;
 974 }
 975
 976 static int blkfront_is_ready(struct xenbus_device *dev)
 977 {
 978         struct blkfront_info *info = dev->dev.driver_data;
 979
 980         return info->is_ready;
 981 }
 982
 983 static int blkif_open(struct inode *inode, struct file *filep)
 984 {
 985         struct blkfront_info *info = inode->i_bdev->bd_disk->private_data;
 986         info->users++;
 987         return 0;
 988 }
 989
 990 static int blkif_release(struct inode *inode, struct file *filep)
 991 {
 992         struct blkfront_info *info = inode->i_bdev->bd_disk->private_data;
 993         info->users--;
 994         if (info->users == 0) {
 995                 /* Check whether we have been instructed to close.  We will
 996                    have ignored this request initially, as the device was
 997                    still mounted. */
 998                 struct xenbus_device *dev = info->xbdev;
 999                 enum xenbus_state state = xenbus_read_driver_state(dev->otherend);
1000
1001                 if (state == XenbusStateClosing && info->is_ready)
1002                         blkfront_closing(dev);
1003         }
1004         return 0;
1005 }
1006
1007 static struct block_device_operations xlvbd_block_fops =
1008 {
1009         .owner = THIS_MODULE,
1010         .open = blkif_open,
1011         .release = blkif_release,
1012         .getgeo = blkif_getgeo,
1013         .ioctl = blkif_ioctl,
1014 };
1015
1016
1017 static struct xenbus_device_id blkfront_ids[] = {
1018         { "vbd" },
1019         { "" }
1020 };
1021
1022 static struct xenbus_driver blkfront = {
1023         .name = "vbd",
1024         .owner = THIS_MODULE,
1025         .ids = blkfront_ids,
1026         .probe = blkfront_probe,
1027         .remove = blkfront_remove,
1028         .resume = blkfront_resume,
1029         .otherend_changed = backend_changed,
1030         .is_ready = blkfront_is_ready,
1031 };
1032
1033 static int __init xlblk_init(void)
1034 {
1035         if (!is_running_on_xen())
1036                 return -ENODEV;
1037
1038         if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
1039                 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
1040                        XENVBD_MAJOR, DEV_NAME);
1041                 return -ENODEV;
1042         }
1043
1044         return xenbus_register_frontend(&blkfront);
1045 }
1046 module_init(xlblk_init);
1047
1048
1049 static void __exit xlblk_exit(void)
1050 {
1051         return xenbus_unregister_driver(&blkfront);
1052 }
1053 module_exit(xlblk_exit);
1054
1055 MODULE_DESCRIPTION("Xen virtual block device frontend");
1056 MODULE_LICENSE("GPL");
1057 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
1058 MODULE_ALIAS("xen:vbd");
1059 MODULE_ALIAS("xenblk");